| Unique #'s | Mean | SD | 25th %ile | Median | 75th %ile | Max | |
|---|---|---|---|---|---|---|---|
| Docked | 3 | 0.4 | 0.5 | 0 | 0 | 1 | 2 |
| Dockless | 5 | 0.5 | 0.7 | 0 | 0 | 1 | 5 |
| E-Scooter | 10 | 1 | 2 | 0 | 1 | 2 | 10 |
The Effect of Micromobility on DUIs
Goucher College
Old Dominion University
University of Alabama
Rensselaer Polytechnic Institute
In general, transportation in the United States is dominated by automobile. However, recent innovations (e.g. rideshare) have provided alternative options that may reduce vehicle dependency. One of the most recent of these innovations is micromobility.
. . .
According to the Bureau of Transportation Statistics, micromobility options have been in about 400 “cities” across the United States.
The effects of alternative transportation options on impaired driving outcomes is a topic that has received quite a bit of attention (Fell et al. 2020). Rideshare, in particular, accounts for a significant portion of this interest over the past half-decade.
Little research exists about micromobility, however. Button, Frye, and Reaves (2020) is, to our knowledge, the only micromobility paper published in an economics journal.
. . .
That said, the two closest papers to our topic are:
In this study, we ask how the introduction of micromobility impacts DUI arrests. We hypothesize that, like rideshare, micromobility could provide an alternative to driving (an automobile) under the influence.
| Unique #'s | Mean | SD | 25th %ile | Median | 75th %ile | Max | |
|---|---|---|---|---|---|---|---|
| Docked | 3 | 0.4 | 0.5 | 0 | 0 | 1 | 2 |
| Dockless | 5 | 0.5 | 0.7 | 0 | 0 | 1 | 5 |
| E-Scooter | 10 | 1 | 2 | 0 | 1 | 2 | 10 |
| Num. Micromobility Modes | 0 | 1 | 2 | 3 |
|---|---|---|---|---|
| Num. Cities/Agencies | 191 | 134 | 53 | 6 |
| Mode | Docked | Dockless | E-Scooters | Total |
|---|---|---|---|---|
| Docked | 0.7 | 2.1 | 141 | |
| Dockless | 23.2 | 26.8 | 138 | |
| E-Scooters | 21.4 | 19.2 | 229 |
In our estimation sample, we only consider agencies:
For mode-specific estimations we only consider agencies exposed to that mode first, and drop observations of control agencies once they obtain other types of micromobility. We can also drop observations of agencies following firm exits, but our results do not change much.
We begin by estimating a Poisson regression of the form:
\[\text{DUI}_{at} = \color{red}{\delta} M_{at} + X_{at}\beta + \alpha_a + \tau_t + \epsilon_{at}\]
We also use the new Local Projections Difference-in-Differences (Dube et al. 2023) estimator to investigate the effects of micromobility.
\[\text{DUI}_{a,t+h} - \text{DUI}_{a,t-1} = \color{red}{\delta_h^{LP}}\Delta M_{a,t} + \tau_{t}^{h} + e_{at}^{h}\]
using only observations where \(\Delta M_{at} = 1\) (newly treated) or \(M_{a,t+h} = 0\) (clean control). This is akin to the “stacking” estimator in Cengiz et al. (2019).
We estimate a \(\approx\) 10% reduction in DUIs per month due to the introduction of micromobility.
While this may seem like an attractive headline, the interpretation is delicate and unlike that of rideshare. A reduction of DUIs indicates less drunk driving, but the actual safety implications are much less clear.
Southern Economic Association (2024)